Steven's Myth of Verizon AMPS coverage in the San Francisco BayArea

Todd Allcock wrote:

> For example, if the survey was a question like "do you get your news from
> TV, radio, internet or magazines?" asking a group comprised entirely of
> magazine subscribers would obviously skew the results.

I guess that John is trying to convince people that if somehow you could
get a sample size of 50,000 respondents, and it was all random, that the
results would be different. Of course this is ridiculous, the sample of
CR subscribers that are Verizon subscribers, are not going to be biased
for or against Verizon, any more than the Sprint, Cingular or T-Mobile
subscribers are going to be biased against their own carriers. I think
what he doesn't understand, is that the survey isn't asking 50,000
people "which carrier is best in your city?," it's asking for an
evaluation of your own carrier.

Now if you surveyed only long distance truck drivers, and trucking
firms, of course Verizon is going to have a huge advantage, because they
have much wider coverage than Cingular in non-urban areas due to AMPS.
The commercial carriers still use AMPS in areas where there is no CDMA
coverage (see "http://www.etrucker.com/apps/news/article.asp?id=51944").

> But asking "cable TV subscribers," "Ford automobile owners," or
> "bricklayers" objective questions about cellular service should tend to
> get the same results if the sample sizes are large enough. (Unless, for
> example, cell companies discriminate against bricklayers...)

Yes, that's the whole point.

Of course Navas is just extremely upset that for yet another year,
Cingular fared extremely poorly in the Consumer Reports survey AND the
J.D. Power surveys. That's why he feels compelled to make up ridiculous
stories to try and defend them, part of which is trying to attack the
companies doing the surveys.

> Perhaps, but that old "standard deviation" equation tends to insure work
> things out. ;-)

The margin of error is still extremely small, even when broken down by
region and then by carrier. Additionally, in some regions, such as the
San Francisco Bay Area, there is such a large difference, that even with
the maximum amount of error applied, Cingular still does extremely
poorly, and Verizon does extremely well.


[Copied to alt.cellular.attws. Please post all alt.cellular.cingular
posts to alt.cellular.attws as well. The Cingular name is going away,
and alt.cellular.attws is the proper venue for posts regarding AT&T's
Wireless Service.]

On Mon, 12 Feb 2007 16:52:16 -0800, SMS <scharf.steven@geemail.com>
wrote in <45d10bc3$0$27176$742ec2ed@news.sonic.net>:

>Todd Allcock wrote:
>
>> For example, if the survey was a question like "do you get your news from
>> TV, radio, internet or magazines?" asking a group comprised entirely of
>> magazine subscribers would obviously skew the results.
>
>I guess that John is trying to convince people that if somehow you could
>get a sample size of 50,000 respondents, and it was all random, that the
>results would be different.

I'm not trying to convince anyone of anything -- I'm just stating the
science of statistics -- makes no difference how many samples you take
when the sampling isn't random, as in the case of CR, where is the
population isn't representative, and the sample is self-selected, two
serious flaws.

>Of course this is ridiculous, the sample of
>CR subscribers that are Verizon subscribers, are not going to be biased
>for or against Verizon, any more than the Sprint, Cingular or T-Mobile
>subscribers are going to be biased against their own carriers. I think
>what he doesn't understand, is that the survey isn't asking 50,000
>people "which carrier is best in your city?," it's asking for an
>evaluation of your own carrier.

You clearly don't understand the concept of sample bias, which has
nothing to do with your kind of bias.

>Now if you surveyed only long distance truck drivers, and trucking
>firms, of course Verizon is going to have a huge advantage, because they
>have much wider coverage than Cingular in non-urban areas due to AMPS.
>The commercial carriers still use AMPS in areas where there is no CDMA
>coverage (see "http://www.etrucker.com/apps/news/article.asp?id=51944").

Again, that's irrelevant, because that's not how sampling works.

>Of course Navas is just extremely upset that for yet another year,
>Cingular fared extremely poorly in the Consumer Reports survey AND the
>J.D. Power surveys. That's why he feels compelled to make up ridiculous
>stories to try and defend them, part of which is trying to attack the
>companies doing the surveys.

Nothing of the sort. I have no agenda -- I'm simply sick of your
agenda. In fact the differences in the better surveys (e.g., J.D.
Powers) are actually small, probably within the level of sampling error.

>> Perhaps, but that old "standard deviation" equation tends to insure work
>> things out. ;-)
>
>The margin of error is still extremely small, even when broken down by
>region and then by carrier.

There'a actually no way to know the actual margin of error, given the
way the survey was taken, and the small sample size when broken down by
area makes it even worse.

>Additionally, in some regions, such as the
>San Francisco Bay Area, there is such a large difference, that even with
>the maximum amount of error applied, Cingular still does extremely
>poorly, and Verizon does extremely well.

Again, the differences in the better surveys are actually small,
probably within the level of sampling error.

--
Best regards, FAQ FOR CINGULAR WIRELESS:
John Navas <http://en.wikibooks.org/wiki/Cingular_Wireless_FAQ>

At 13 Feb 2007 01:26:47 +0000 John Navas wrote:

> when the sampling isn't random, as in the case of CR, where is the
> population isn't representative, and the sample is self-selected, two
> serious flaws.

Non-random and non-representative are not the same thing, necessarily.
If I wanted to determine the average diameter of M&M candies, I could
open a bag, and average the diameters of all of the blue ones. While the
sample was non-random, and perhaps flawed statistically, the sample was
certainly representative and I'll guarantee you my "real world" answer
would be
correct within the margin of error!

Same with the CR study- as non-random as the sample might have been,
there seems to be no good reason to believe why it would not be
representative.

Todd Allcock wrote:

> Same with the CR study- as non-random as the sample might have been,
> there seems to be no good reason to believe why it would not be
> representative.

You could probably find a group of people that was non-representative of
the population as a whole, i.e., heavy urban users such as real estate
agents, highly mobile users with a lot of non-urban use such as
truckers, highly mobile users with mainly urban use such as airline
pilots and flight attendants, etc. It might actually be a useful metric
to know which carriers these groups favor and why.

However nothing suggests that CR subscribers are not representative of
the population as a whole. They are generally higher income, and of
higher education level, which means that they travel more, but this
makes the CR survey even more valuable, for those that are interested in
the best coverage.

I think that we all understand that it's all a sour grapes issue by
Navas. On the plus side, the digression into coffee was very interesting.

On Tue, 13 Feb 2007 00:51:15 -0700, Todd Allcock
<elecconnec@AmericaOnLine.com> wrote in <eqrqm0$q2b$1@aioe.org>:

>At 13 Feb 2007 01:26:47 +0000 John Navas wrote:
>
>> when the sampling isn't random, as in the case of CR, where is the
>> population isn't representative, and the sample is self-selected, two
>> serious flaws.
>
>Non-random and non-representative are not the same thing, necessarily.
>If I wanted to determine the average diameter of M&M candies, I could
>open a bag, and average the diameters of all of the blue ones. While the
>sample was non-random, and perhaps flawed statistically, the sample was
>certainly representative and I'll guarantee you my "real world" answer
>would be
>correct within the margin of error!
>
>Same with the CR study- as non-random as the sample might have been,
>there seems to be no good reason to believe why it would not be
>representative.

The good reason is that CR subscribers have considerably different
demographics than the universe of cellular, making any CR survey
unrepresentative. Then there's the problem of self-selction.

--
Best regards, FAQ FOR CINGULAR WIRELESS:
John Navas <http://en.wikibooks.org/wiki/Cingular_Wireless_FAQ>

At 15 Feb 2007 17:23:52 +0000 John Navas wrote:

> The good reason is that CR subscribers have considerably different
> demographics than the universe of cellular, making any CR survey
> unrepresentative.

So that begs the question is the CR subscribers' demographic more likely
or less likely to have good service or bad service? I can't personally
envision a scenario that would cause CR subscribers to have vastly
different cell service than non-subscribers.

> Then there's the problem of self-selction.

I'll agree that's a bigger problem, assuming that unhappy customers might
be more likely to "grumble" about service than happy ones are to gush.
But again, that would drag everyone's scores down- not one carrier's, so
the relative results wouldn't be much different.

I suppose the easiest way to make the study accurate for me, would be
simply to subscribe to CR, then the results would apply to me, since I
would be part of the represented universe.

I'll let you all know how my subscription to CR affects my cellphone
reception... ;-)

On Thu, 15 Feb 2007 14:15:19 -0700, Todd Allcock
<elecconnec@AmericaOnLine.com> wrote in <er2lbh$drb$3@aioe.org>:

>At 15 Feb 2007 17:23:52 +0000 John Navas wrote:
>
>> The good reason is that CR subscribers have considerably different
>> demographics than the universe of cellular, making any CR survey
>> unrepresentative.
>
>So that begs the question is the CR subscribers' demographic more likely
>or less likely to have good service or bad service?

With all due respect, there is no such question -- you can only make
valid generalizations from proper sampling, and the CR survey isn't even
a proper sampling of CR subscribers (because of self-selection).
There's simply no way to know the sample bias. It's thus just pure
speculation.

>I can't personally
>envision a scenario that would cause CR subscribers to have vastly
>different cell service than non-subscribers.

I can easily think of lots of things that would bias the result in
unpredictable ways, but without real data, there's simply no way of
knowing which of them might be at work.

>> Then there's the problem of self-selction.
>
>I'll agree that's a bigger problem, assuming that unhappy customers might
>be more likely to "grumble" about service than happy ones are to gush.
>
>But again, that would drag everyone's scores down- not one carrier's, so
>the relative results wouldn't be much different.

There are lots of other possible factors that might well differ from
carrier to carrier, but again, without real data it's all just
speculation.

>I suppose the easiest way to make the study accurate for me, would be
>simply to subscribe to CR, then the results would apply to me, since I
>would be part of the represented universe.

1. That's not how statistics works. The issue is learning something
about the entire target universe, not any one individual.

2. That wouldn't solve the problem of self-selection.

>I'll let you all know how my subscription to CR affects my cellphone
>reception... ;-)

Whatever.

--
Best regards, FAQ FOR CINGULAR WIRELESS:
John Navas <http://en.wikibooks.org/wiki/Cingular_Wireless_FAQ>

Todd Allcock wrote:

> So that begs the question is the CR subscribers' demographic more likely
> or less likely to have good service or bad service? I can't personally
> envision a scenario that would cause CR subscribers to have vastly
> different cell service than non-subscribers.

Even though there are certainly some demographic differences between CR
subscribers and the general public, none of these would materially
affect the results of one carrier more than another carrier. The
demographics of Consumer Reports readers versus the general population
include higher education level, higher income, more moderate
politically, and middle-age to older age. On the one hand, these
demographics would tend to make people more critical of products and
services, but on the other hand, these demographics may understand the
limitations of cellular communications better than uneducated people, so
they may cut the carriers more slack.

> I'll agree that's a bigger problem, assuming that unhappy customers might
> be more likely to "grumble" about service than happy ones are to gush.
> But again, that would drag everyone's scores down- not one carrier's, so
> the relative results wouldn't be much different.

I read a funny article about that in SmartMoney magazine yesterday. It
seems that there is very little middle ground in most reviews by
individuals of products and services. They either hate something or love
it. I.e. a five star review for a stapler at Amazon, "This stapler is
great. It works very well and staples many papers together."

What makes the CR survey so valuable is that they don't just ask "how's
your service?" they go into great detail with questions on specific
aspects of the service. But as you stated, even if the people that
choose to respond are more or less critical of a service, this will
extend across all the carriers, and will cancel out.

In earlier years, Navas complained that the reason Cingular was rated so
poorly was that their TDMA subscribers were dragging down their scores
(though in reality the TDMA/AMPS subscribers were probably dragging the
scores up). Now that the TDMA subscribers are an insignificant portion
of the total, he's come up with a new excuse, apparently he believes
that the Cingular subscribers that choose to respond to the survey are
somehow more critical of their service than the Verizon subscribers that
choose to respond. Certainly if there were any evidence that CR's
subscribers demographics somehow benefit one carrier over another, he'd
have presented that evidence, or at least a theory of that evidence. As
you pointed out "non-random and non-representative are not the same thing."

> I suppose the easiest way to make the study accurate for me, would be
> simply to subscribe to CR, then the results would apply to me, since I
> would be part of the represented universe.
>
> I'll let you all know how my subscription to CR affects my cellphone
> reception... ;-)

I read it at the library, so I guess I'm not represented in the survey
either.

On Thu, 15 Feb 2007 16:09:59 -0800, SMS <scharf.steven@geemail.com>
wrote in <45d4f65d$0$27188$742ec2ed@news.sonic.net>:

>Todd Allcock wrote:
>
>> So that begs the question is the CR subscribers' demographic more likely
>> or less likely to have good service or bad service? I can't personally
>> envision a scenario that would cause CR subscribers to have vastly
>> different cell service than non-subscribers.
>
>Even though there are certainly some demographic differences between CR
>subscribers and the general public, none of these would materially
>affect the results of one carrier more than another carrier.

There's simply no way of knowing that without real data, which doesn't
exist, so that's just your own unsupported speculation.

>The
>demographics of Consumer Reports readers versus the general population
>include higher education level, higher income, more moderate
>politically, and middle-age to older age.

Citation? Apparently not.

>On the one hand, these
>demographics would tend to make people more critical of products and
>services, but on the other hand, these demographics may understand the
>limitations of cellular communications better than uneducated people, so
>they may cut the carriers more slack.

Again, just your speculation.

>What makes the CR survey so valuable is that they don't just ask "how's
>your service?" they go into great detail with questions on specific
>aspects of the service.

Actually, it doesn't -- the survey is based on simple generic multiple
choice questions (as you should know if you had any actual knowledge of
the methodology).

>But as you stated, even if the people that
>choose to respond are more or less critical of a service, this will
>extend across all the carriers, and will cancel out.

Again, just your speculation.

>In earlier years, Navas complained that the reason Cingular was rated so
>poorly was that their TDMA subscribers were dragging down their scores
>(though in reality the TDMA/AMPS subscribers were probably dragging the
>scores up).

What I actually did was point out the invalidity of lumping multiple
types of GSM and TDMA together, just like the invalidity of lumping
together Sprint CDMA2000 and Nextel iDEN.

>Now that the TDMA subscribers are an insignificant portion
>of the total, he's come up with a new excuse, apparently he believes
>that the Cingular subscribers that choose to respond to the survey are
>somehow more critical of their service than the Verizon subscribers that
>choose to respond.

I've said nothing of the kind. What I have done is point out the
invalidity of the CR survey. Regardless, as I expected, Cingular is
doing much better indeed in the market now that most of the migrations
are behind it.

>Certainly if there were any evidence that CR's
>subscribers demographics somehow benefit one carrier over another, he'd
>have presented that evidence, or at least a theory of that evidence.

On the contrary -- there's simply no way to know the sampling bias
without real data, which simply doesn't exist.

>As
>you pointed out "non-random and non-representative are not the same thing."

I guess you have no idea how meaningless that is.

--
Best regards, FAQ FOR CINGULAR WIRELESS:
John Navas <http://en.wikibooks.org/wiki/Cingular_Wireless_FAQ>

John Navas <spamfilter1@navasgroup.com> wrote in
news:aqn9t294p65unfp3hpt9maatd4oi5p58eo@4ax.com:

> On Thu, 15 Feb 2007 14:15:19 -0700, Todd Allcock
> <elecconnec@AmericaOnLine.com> wrote in <er2lbh$drb$3@aioe.org>:
>
>>At 15 Feb 2007 17:23:52 +0000 John Navas wrote:
>>
>>> The good reason is that CR subscribers have considerably different
>>> demographics than the universe of cellular, making any CR survey
>>> unrepresentative.
>>
>>So that begs the question is the CR subscribers' demographic more
likely
>>or less likely to have good service or bad service?
>
> With all due respect, there is no such question -- you can only make
> valid generalizations from proper sampling, and the CR survey isn't
even
> a proper sampling of CR subscribers (because of self-selection).
> There's simply no way to know the sample bias. It's thus just pure
> speculation.


So you are saying that you can't prove any sample bias. Thanks for
playing.


>
>>I can't personally
>>envision a scenario that would cause CR subscribers to have vastly
>>different cell service than non-subscribers.
>
> I can easily think of lots of things that would bias the result in
> unpredictable ways, but without real data, there's simply no way of
> knowing which of them might be at work.


Bullshit. Unless you can prove that CR subscribers have a skewed view
of service and coverage, it is a very real-world sample.


>
>>> Then there's the problem of self-selction.
>>
>>I'll agree that's a bigger problem, assuming that unhappy customers
might
>>be more likely to "grumble" about service than happy ones are to gush.
>>
>>But again, that would drag everyone's scores down- not one carrier's,
so
>>the relative results wouldn't be much different.
>
> There are lots of other possible factors that might well differ from
> carrier to carrier, but again, without real data it's all just
> speculation.
>

You keep talking about "other factors"- there are none. Either they
have the same view of service or they don't. Period.


>>I suppose the easiest way to make the study accurate for me, would be
>>simply to subscribe to CR, then the results would apply to me, since I
>>would be part of the represented universe.
>
> 1. That's not how statistics works. The issue is learning something
> about the entire target universe, not any one individual.
>
> 2. That wouldn't solve the problem of self-selection.

All surveys are self-selected. No survey is done where the respondent
is required to answer questions. All those surveyed make the conscious
decision to participate in the survey, and therefore self-select.

>
>>I'll let you all know how my subscription to CR affects my cellphone
>>reception... ;-)
>
> Whatever.
>

What? He can't use your argument against you?